Modeling the Sources of the 2018 Palu, Indonesia, Tsunami Using Videos From Social Media

Abstract

The 2018 Palu tsunami contributed significantly to the devastation caused by the associated MW 7.5 earthquake. This began a debate about how the moderate size earthquake triggered such a large tsunami within Palu Bay, with runups of more than 10 m. The possibility of a large component of vertical coseismic deformation and submarine landslides have been considered as potential explanations. However, scarce instrumental data have made it difficult to resolve the potential contributions from either type of source. We use tsunami waveforms derived from social media videos in Palu Bay to model the possible sources of the tsunami. We invert InSAR data with different fault geometries and use the resulting seafloor displacements to simulate tsunamis. The coseismic sources alone cannot match both the video‐derived time histories and surveyed runups. Then we conduct a tsunami source inversion using the video‐derived time histories and a tide gauge record as inputs. We specify hypothetical landslide locations and solve for initial tsunami elevation. Our results, validated with surveyed runups, show that a limited number of landslides in southern Palu Bay are sufficient to explain the tsunami data. The Palu tsunami highlights the difficulty in accurately capturing with tide gauges the amplitude and timing of short period waves that can have large impacts at the coast. The proximity of landslides to locations of high fault slip also suggests that tsunami hazard assessment in strike‐slip environments should include triggered landslides, especially for locations where the coastline morphology is strongly linked to fault geometry.

New High-Resolution Modeling of the 2018 Palu Tsunami, Based on Supershear Earthquake Mechanisms and Mapped Coastal Landslides, Supports a Dual Source

“…All three sources, however, significantly underpredict runups observed in the southern part of the bay (south of ∼ − 0.75 + N), with a maximum of 5 m whereas observed runups reached up to 10.5 m. In contrast, runups are relatively well predicted in the northern part of the Bay by Jamelot et al (2019)'s and Ulrich et al (2019)'s coseismic sources, with a slight advantage for the latter. This agrees with conclusions of earlier studies that coseismic sources alone cannot explain the tsunami coastal impact, particularly in the south (e.g., Nakata et al, 2020;Sepúlveda et al, 2020). Consistent with these results, Figure 4 shows that measured or inferred time series of surface elevation are not well reproduced, particularly in the southern part of the bay.…”

“…shorter period waves generated by the coastal landslide were followed by longer period waves from the earthquake. Pakoksung et al (2019), Nakata et al (2020), and Sepúlveda et al (2020) identified and modeled landslides as the most important, if not the principal, contributors to the tsunami. Their landslide parameters and corresponding wave generation, however, were hypothetical and selected to match observations.…”

“…200 m resolution grid with two nested 10 m grids in palu city and pantoloan, derived from DEMNAS and BATNAS USGS (2018) and their own (hybrid source with nine subfaults parameterized from satellite data), using Okada (1985) and Tanioka and Satake (1996) (Sepúlveda et al, 2020). Here, as noted by Carvajal et al (2019), Sepúlveda et al (2020), and Liu et al (2020), the tide gauge is located in shallow water inside a harbor basin protected by a slotted seawall/dock (see Supplementary Figure S1 in supplementary material), which is not represented in the available bathymetric data nor in our model grids. While not affecting tide measurements, harbor structures may cause seiching and affect tsunami wave dynamics by modifying their elevation through reflection or dissipation; later in time ( > 650 s), the record may have been affected by waves reflected from the other side of the bay.…”

“…While not affecting tide measurements, harbor structures may cause seiching and affect tsunami wave dynamics by modifying their elevation through reflection or dissipation; later in time ( > 650 s), the record may have been affected by waves reflected from the other side of the bay. The 1 Hz tide gauge measurements were averaged over 30 s and provided only every minute (BIG, 2018;Sepúlveda et al, 2020), yielding the fairly coarse time series plotted in Figure 4B. For this reason, while the gauge accurately measured long period waves, shorter waves, such as from landslides, were not recorded.…”

“…Although there are now many tsunami simulations of the 2018 Palu event (e.g., Heidarzadeh et al, 2019;Takagi et al, 2019;Carvajal et al, 2019;Pakoksung et al, 2019;Gusman et al, 2019;Jamelot et al, 2019;Ulrich et al, 2019;Goda et al, 2019;Nakata et al, 2020;Sepúlveda et al, 2020;Liu et al, 2020, see summary of studies characteristics in Table 1), the tsunami mechanism, earthquake, coastal landslides, or both in combination, is still uncertain. In addition, whereas many published models simulate some, or even most, recorded runups around the Bay, the mechanisms are often ad hoc and do not reproduce the timing of tsunami waves from eyewitness accounts or the video evidence.…”